ZeST / DPT /EVALUATION.md
fffiloni's picture
Upload 47 files
a9289c0
|
raw
history blame
3.59 kB

Genral-purpose models

The general-purpose models are affine-invariant and as such need a pre-alignment step before an error can be computed.

Sample code for NYUv2 can be found here: https://gist.github.com/ranftlr/a1c7a24ebb24ce0e2f2ace5bce917022

Sample code for KITTI can be found here: https://gist.github.com/ranftlr/45f4c7ddeb1bbb88d606bc600cab6c8d

KITTI

Result:

Evaluating 697 files
GT files reading done
45 GT files missing
Computing errors
     d1,      d2,      d3,  AbsRel,   SqRel,    RMSE, RMSElog,   SILog,   log10
  0.959,   0.995,   0.999,   0.062,   0.222,   2.575,   0.092,   8.282,   0.027
Done.

NYUv2

Result (old model) - from paper:

Evaluating 654 files
GT files reading done
0 GT files missing
Computing errors
     d1,      d2,      d3,  AbsRel,   SqRel,    RMSE, RMSElog,   SILog,   log10
  0.904,   0.988,   0.998,   0.109,   0.054,   0.357,   0.129,   9.521,   0.045
Done.

Result (new model):

GT files reading done
697 GT files missing
Computing errors
     d1,      d2,      d3,  AbsRel,   SqRel,    RMSE, RMSElog,   SILog,   log10
  0.905,   0.988,   0.998,   0.109,   0.055,   0.357,   0.129,   9.427,   0.045
Done.